What AI visibility does and does not send you
The category is sold on visibility, and visibility is real. But the traffic arithmetic behind it is unlike any channel you have bought before, and getting it wrong is how a working programme gets cancelled at month six.
Mentions and clicks are separate outcomes
When an assistant recommends you inside an answer, three things can happen: the user clicks a citation, the user searches your brand name separately, or the user simply remembers the name and arrives days later by some other route. Only the first is directly attributable. The other two are real effects that will appear in your data as branded search or direct traffic — channels nobody credits to this work.
The practical consequence: measure AI referral clicks, but do not judge the programme on them alone. Watch branded search volume and direct traffic over the same period. A programme that is working often shows a modest referral number alongside a clearer lift in people arriving already knowing who you are.
Instrument the clicks you can see
Assistant referrals do arrive with identifiable referrer information, and most analytics setups bucket them into “referral” or lose them in direct. Two hours of setup makes them visible:
- Build a channel group or segment for the assistant referrers you care about, and check it quarterly — new surfaces appear and old ones change their referrer behaviour.
- Report that traffic separately from search. It behaves differently: fewer sessions, later in the buying process, and typically a higher conversion rate than cold organic. Blending it into organic hides both facts.
- Ask on your forms. A “how did you hear about us” field is unfashionable and it is the only instrument that catches the two invisible paths above.
Set expectations in the right order
Judge the first quarter on whether you are being mentioned for the prompts you chose, which is measurable from your own prompt log. Judge the second on whether mentions are accurate and favourable — being described wrongly is worse than being absent. Judge the third on whether any of it is producing conversations. Agencies that promise the third outcome in the first quarter are describing a timeline nobody controls.
The number to watch that nobody sells you
Accuracy. Run your prompt set and read what the assistants actually say about your pricing, your service area, your product lineup, and your competitors' relative strengths. Wrong information stated confidently to a buyer costs more than absence does, and it usually traces back to a stale page, an old press release, or a directory listing nobody has updated in three years — all cheap to fix, and none of it requiring a retainer to find.